Here's a trivial example to illustrate the difference between pattern recognition and reasoning and it impacts behavior generation: let's say you encounter, for the very first time, a glass door with ⅃⅃Uꟼ written on it.
Show this thread
-
Pattern recognition: nearest neighbor is "PULL", I've learned to associate that with pulling the door. I pull. Reasoning: that's a mirror image of "PULL". Must be written on the other side. The door will open by pulling towards the other side, i.e. pushing from this side. I push
Show this thread
Only multi-step reasoning (which is slow and resource-intensive) enables zero-shot adaptation to novel situations. Meanwhile, pattern recognition makes navigating well-known situations faster and more efficient.
-
-
System 1 vs system 2. Problem w/ AI: – reward: learn a posteriori; for novel situations: random – reasoning: programmed a priori; not feasible for Zipf novelty either To learn to reason, & reason to learn (experiment) AI needs the upper level to semantics: purpose (cybernetics)
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
This also shows that most modern reinforcement learning agents are missing a critical “thinking” subenvironment where they learn to interact with their own model of the world.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Multi-step reasoning doesn’t have to be slow. Consider analysis-by-synthesis models like Yildirim et al. 2020. Using pattern recognition techniques as function approximators in models that are able to reason is a really exciting avenue.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.